Read our Technical White Paper
Building a Large Scale Machine Learning-Based Anomaly Detection System
Part 1: Design Principles
Wix.com is a leading cloud-based web development platform with millions of users worldwide. The company’s mission is to make it simple for everyone to create a beautiful, professional web presence. No creative limits, no coding skills are needed to get businesses online. The company is headquartered in Tel Aviv, with offices in San Francisco, New York City, Dnipropetrovsk, and Vilnius.
As a subscription based web service, retention of customers for Wix is critical. This means that analysts typically continuously monitor metrics that indicate important aspects of user activity, to understand if customers are successfully using their solution. This is typical of many Web businesses such as subscription-based applications and ecommerce sites.
Before using Anodot, vast amounts of data were measured and analyzed manually by data analysts. This could include activity related to customers actions on Wix – e.g. success and failure rates while opening the Wix website editors – checkouts at e-commerce sites hosted by Wix, logins by premium customers, and other important events. Analysts spent a great deal of time scrutinizing reports and graphs manually to try to detect issues, but important issues were sometimes identified hours to days after they had occurred.
Wix needed a real-time alert system that would indicate issues without manual threshold settings in the key metrics. Anodot proved to be the system required for providing the necessary insights to the company’s analysts.
Anodot receives a feed of the same metrics that the analysts had been gathering all along – around 450 million events per day – and initially used this feed to determine a normal baseline. Once the normal was determined automatically through proprietary machine learning algorithms, Anodot can now immediately identify abnormal events, whether in one metric or across several.
Prior to Anodot, Wix’s team was tracking just a handful of metrics, and always with a significant delay in analysis. Now that they have Anodot, more than 80 Wix analysts have configured 75+ dashboards in Anodot, and scores of anomaly alerts. In addition to the configured alerts, Anodot can surface anomalies in metrics that had never been tracked before.
By using Anodot, Wix’s business intelligence, R&D and devops teams now have a single, unified platform for understanding what is happening on their site. They are able to identify anomalies quickly, and achieve a more rapid root cause analysis for any issue that comes up.
“With Anodot we are able to detect very early and change decisions that have direct impact on our key business activities. We can also investigate issues in minutes, not hours. It does its magic every day, with little effort on our side. Anodot has become a must have solution across a variety of our groups – BI, R&D and the DevOps team.”
Mark Sonis, Wix